Meta-Eval/EvalHarness
3
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Narration lines all plain white (color scheme toggle kept in _phase_color for quick restore)

Browse Source 
Co-Authored-By: Claude <noreply@anthropic.com>
This commit is contained in:
sora 2026-09-10 11:19:58 +00:00
parent 5734f7fbe6
commit 5b595f3a10
1 changed files with 13 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
evalharness/cli.py
View File

@ -199,10 +199,11 @@ def _print_run_plan(console, args, model_spec):
def _phase_color(message: str) -> str: def _phase_color(message: str) -> str:
"""Start/finish color code: events that START or are IN progress are """Narration line color. Currently PLAIN WHITE for everything (user
yellow; events that COMPLETE are green. One glance = what stage, preference); flip the returns to 'yellow'/'green' to restore the
did it finish.""" start/finish coloring scheme."""
m = message.lower() return ''
m = message.lower() # unreachable, kept for quick restore
if any(k in m for k in ('complete', 'ready', 'downloaded', 'parsed', if any(k in m for k in ('complete', 'ready', 'downloaded', 'parsed',
'restored', 'ok (')): 'restored', 'ok (')):
return 'green' return 'green'
@ -215,7 +216,10 @@ def _print_phase(console, index, total, name, message):
text = f'{prefix}{name}: {message}' text = f'{prefix}{name}: {message}'
color = _phase_color(message) color = _phase_color(message)
if console is not None: if console is not None:
if color:
console.print(f'[{color}]{text}[/{color}]', highlight=False) console.print(f'[{color}]{text}[/{color}]', highlight=False)
else:
console.print(text, highlight=False)
else: else:
print(text, flush=True) print(text, flush=True)
@ -493,7 +497,8 @@ def _cmd_eval_run(args) -> int:
def _emit(msg, _i=i, _n=name): def _emit(msg, _i=i, _n=name):
if _shared_reporter is not None: if _shared_reporter is not None:
_col = _phase_color(msg) _col = _phase_color(msg)
_shared_reporter.log(f'[{_col}][{_i + 1}/{total_runs}] {_n}: {msg}[/{_col}]') _line = f'[{_i + 1}/{total_runs}] {_n}: {msg}'
_shared_reporter.log(f'[{_col}]{_line}[/{_col}]' if _col else _line)
else: else:
_print_phase(console, _i + 1, total_runs, _n, msg) _print_phase(console, _i + 1, total_runs, _n, msg)
@ -531,7 +536,8 @@ def _cmd_eval_run(args) -> int:
_tag = f'[{_idx}/{total_runs}] ' if total_runs > 1 else '' _tag = f'[{_idx}/{total_runs}] ' if total_runs > 1 else ''
if _reporter is not None: if _reporter is not None:
_col = _phase_color(msg) _col = _phase_color(msg)
_reporter.log(f'[{_col}]{_tag}{_name}: {msg}[/{_col}]') _line = f'{_tag}{_name}: {msg}'
_reporter.log(f'[{_col}]{_line}[/{_col}]' if _col else _line)
if 'scoring' in msg: if 'scoring' in msg:
_reporter.set_phase('scoring') _reporter.set_phase('scoring')
elif 'generating model responses' in msg: elif 'generating model responses' in msg: